MS MS - Management Station MNE - Managed Network Element. Link l LAN MNE MNE...

Analyzing Mobile Agent Scalability in Network

Management

Marcelo Gon alves Rubinstein and Otto Carlos Muniz Bandeira Duarte

{rubi, otto}@gta.ufrj.br

Grupo de Teleinformtica e Automa o - GTA

Universidade Federal do Rio de Janeiro

COPPE/EE - Programa de Engenharia Eltrica

P.O. Box 68504 - 21945-970 - Rio de Janeiro - RJ - Brazil

Abstract

This paper aims to investigate mobile agent scalability in network management, in order to
nd when mobile agents can improve the management eciency. Mobile agent performance is compared with the SNMP one in management task simulations. Response time results show that the mobile agent is less sensitive to the latency and the bandwidth of a bottleneck link that connects the management station to a remote LAN, but the mobile agent is more inuenced by the task to be performed. The results also show that the mobile agent performs better when the number of managed network elements ranges between two limits: an inferior bound, associated with the number of messages that traverse the bottleneck link, and a superior bound, related to the incremental size of the mobile agent, which turns migration dicult. Moreover, the performance increases when the agent returns to the management station after visiting a
xed number of nodes.

1 Introduction

Agents are programs that help users to perform tasks, acting on behalf of these users. If these agents can execute their tasks on the network, they are called mobile agents. Mobile agents can move to the place where data are stored and select information the user wants by using intelligence saving bandwidth, time, and money.

Network management can benet from the use of mobile agents, due to the fact that, in a general manner, this management is based on a centralized paradigm. SNMP (Simple Network Management Protocol) 1] and CMIP (Common Management Information Protocol) 2] use the

client-server model, on which the manager centralizes information and gives orders to perform repair actions, and the agent collects and saves information in a database and executes manager orders. This centralized approach does not scale when the size or the complexity of the network increases and may not work well when there is a point of failure on the network or in the manager. Mobile agents decentralize processing and control, and, as a consequence, reduce the trac around the management station, turn asynchronous agent-manager communication (useful when there are unreliable or lossy links), distribute processing load, and increase the exibility of the management agents' behavior.

Research activities related to mobile code in network management are still recent 3]. Man-agement by Delegation (MbD) 4] has been the rst paradigm to address decentralization and automation of management tasks by dynamically delegating management functions to agents. Mobile agent applications in the OSI functional areas of network management are being de-veloped by Pagurek et al. 5]. They have implemented a code mobility infrastructure and a suite of simple tools that interact with agents located on network components. Baldi et al. 6] evaluate the tradeos of mobile code design paradigms in network management applications by developing a quantitative model for traditional and mobile code design of management func-tionalities. Morin et al. 7] implement a management architecture that consists of a server and several managers communicating with SNMP agents located on the managed network elements. They perform measurements of bandwidth utilization for this architecture on an Ethernet LAN, while Rubinstein and Duarte 8] simulate management tasks performed by mobile agents and SNMP ones, comparing both the approaches, on topologies similar in shape to the Internet. This paper aims to investigate mobile agent scalability in network management. We analyze the performance of a management system and compare mobile agent and SNMP approaches. The performance is evaluated for: the latency and the data rate of the link that connects the management station to the managed LAN, the initial size of mobile agent, the task to be performed, and the gain due to the unload procedure of the mobile agent after visiting a xed number of nodes.

This paper is organized as follows. Section 2 describes currently network management systems. Section 3 presents simulation results concerning the applicability of mobile agents in performing management tasks. At last, concluding remarks are presented in Section 4.

2 Network Management Systems

Currently, most network management systems are centralized, but some steps towards decen-tralization have already been taken and new decentralized approaches are also being devel-oped 9].

SNMP and CMIP protocols, proposed by the Internet Engineering Task Force (IETF) and the International Organization for Standardization (ISO), are based on the client-server model, on which the management station acts as a client that provides a user interface to the network manager and interacts with agents, which are servers that manage remote access to their local information.

IETF and ISO have taken some steps towards decentralization. In event notication, SNMP agents notify the management station upon the occurrence of a few signicant events by using traps, i.e., messages sent without an explicit request from the management station. The ISO approach uses more complex agents that have higher processing capacity. In both cases, the agent is only responsible for the notication of an event. SNMPv2 adopts a decentralized approach where there are multiple top-level management stations, the management servers. Each such server is responsible for managing agents, but it can delegate responsibility to an intermediate manager. The intermediate manager, also called proxy agent, plays the role of a manager, in order to monitor and control the agents under its responsibility and it also plays the role of an agent to provide information and accept control from a higher-level management server. SNMPv3 makes additional minor functional changes to the SNMPv2 and incorporates a new security approach 1]. IETF has also proposed RMON (Remote MONitoring) 10] that uses network monitoring devices called monitors or probes. By monitoring packet trac and analyzing the headers, the probes provide information about links, connections among stations, trac patterns, and status of network nodes. A probe can detect failures, misbehaviors, and identify complex events even when not in contact with the management station.

New decentralized architectures are also being developed and some of them use mobile agents to decentralize processing and control from the management station 8].

2.1 Network Management Using Mobile Agents

Mobile agents are created to perform their tasks in dierent network computers. A mobile agent can interrupt its execution and migrate from a machine to another, carrying data about agent's state that include information obtained from previous task executions. As the number of visited nodes grows, mobile agent size also increases, turning migration harder. One possible solution to this problem is to visit a xed number of nodes, return or send all data to the agent home (reducing mobile agent size), and continue the task on the remaining nodes (Section 3.4). The initial size of a mobile agent also aects agent performance because the larger the size, the more dicult the migration (Section 3.2). This initial size depends on the task to be carried out (Section 3.3) and on the language used to implement it.

Lots of advantages 11] may justify agent utilization in network management, such as:

Reduced cost: since management functions require the transfer of a great volume of

data on the network, it may be better to send an agent to perform the task directly on management agents, where data are stored. Thus, the agent can lter and select just the relevant information and transmit them to the management station.

Asynchronous processing: an agent can be sent through the network to perform tasks

on other nodes, and while the agent is out of its home node, this node can be out of operation.

Distributed processing: low capacity computers can be eciently used in order to

per-form simple management tasks, distributing processing previously concentrated on the management station.

Flexibility: a new behavior for the management agents can be determined by the

man-agement station by sending a mobile agent with a new execution code, substituting in real-time the old one.

For the SNMP, network management does not scale when the size or the complexity of the network increases because of the centralized processing and control. Mobile agents can be used in order to solve this problem and it is important to investigate when they improve the management eciency.

3 Simulation Results

The applicability of mobile agents in carrying out network management tasks is assessed by comparing the mobile agent performance with the SNMP one. The topology used in these simulations consists of a LAN of Managed Network Elements (MNEs) connected to the man-agement station by a bottleneck link called l (Figure 1). For the Ethernet LAN, the number of nodes is 250, the data rate is 10 Mbps, and the latency is 10 s. The considered performance parameters are bandwidth consumption on link l, evaluated by the number of bytes transmit-ted and received by a management station, and response time in retrieving SNMP variables (MIB-II objects 12]). MS Link l MNE MNE MNE MNE

. . .

LAN

Figure 1: Topology used in all simulations.

The Network Simulator (NS) 13] is used in this paper. This discrete-event simulator provides several implemented protocols and mechanisms to simulate computer networks with node and link abstractions. In these simulations, we have used UDP and Ethernet implemented functions, but some UDP modules of the NS have had to be modied in order to eectively receive a packet, sending it to the next layer.

The simulation model assumes that links and nodes have no load, links are error-free, and processing time in application layer is not taken into account. UDP is used in all simulations

and UDP and IP headers are also considered. The Maximum Transmission Unit (MTU) used for all links is 1500 bytes, therefore, there is no fragmentation of SNMP messages since they are small. Every request/response of a variable is sent on a dierent message.

SNMP sends requests to all elements to be managed (one after receiving the response from the other) the mobile agent goes to an element to be managed, gathers the variable, and visits all other elements. After nishing, the mobile agent returns to the management station.

Various experiments are carried out in order to evaluate the eect on management performance of: the latency and the data rate of the link l, the initial size of mobile agent, the task to be performed, and the gain due to the unload procedure of the mobile agent after visiting a xed number of nodes.

3.1 Eect of the Latency and Data Rate of the Link

In this section, we evaluate the eect of the latency and data rate of the link l, when retrieving an SNMP variable named ifInErrors, which denotes the number of received packets discarded because of errors, from interfaces of LAN stations. Three dierent values for the latency and the data rate of the link l are used, as in Table 1, and the mobile agent size is 5 kbytes.

Table 1: Values for the latency and the data rate of the link l. Link l Values Latency (ms) Data Rate (Mbps)

v 1 10 0.64 v 2 4 2 v 3 0.1 10 0 1 2 3 4 5 6 1 50 100 150 200 250 Response Time (s)

Number of Managed Network Elements SNMP-Link v1

SNMP-Link v2 SNMP-Link v3

0 1 2 3 4 5 6 1 50 100 150 200 250 Response Time (s)

Number of Managed Network Elements MA-Link v1

MA-Link v2 MA-Link v3

Figure 3: Response time for the mobile agent.

Figure 3 shows that mobile agent's behavior does not change signicantly with the topology, but for the SNMP, the dierence of performance among response times for the three topologies increases faster when the number of MNEs is large (Figure 2), due to the fact that SNMP packets traverse oftentimes the link l. Response time for the SNMP grows approximately linearly with the number of MNEs. For the mobile agent, the response time increases faster when the number of MNEs grows, due to the incremental size of the mobile agent. For the link

v

3 (Table 1), the SNMP performs better than the mobile agent mainly because the link latency

is small, but for the linkv

1, the mobile agent has a smaller response time since the link latency

is large. If the latency is medium (linkv

2), the mobile agent performs better when the number

of MNEs ranges between two limits (Section 3.2).

3.2 Eect of the Initial Size of Mobile Agent

The eect of the initial size of mobile agent is also analyzed for a link l value equal to v 2

(Table 1). Five dierent initial sizes of mobile agent are used (1, 3, 5, 7, and 9 kbytes). The task is the same of the previous experiment.

For a small number of MNEs, the SNMP uses fewer bytes than the mobile agent to perform the task, but as the number of MNEs increases, the overhead associated with several retrievals (PDU GetRequest, UDP, and IP's headers) turns mobile agent utilization more suitable (Fig-ure 4). As mobile agent size increases, the number of bytes associated with the management station increases proportionally. For the response time, as the number of MNEs increases, the performance dierence among the various mobile agent sizes increases faster (Figure 5). Ana-lyzing the curves MA-5kB and SNMP, for a small number of MNEs, the mobile agent performs better, but when the number of MNEs is greater than 200, SNMP performance is better since the mobile agent size is signicant.

0 5 10 15 20 25 30 35 40 1 50 100 150 200 250 Number of kbytes

Number of Managed Network Elements SNMP MA-1kB MA-3kB MA-5kB MA-7kB MA-9kB

Figure 4: Number of bytes for dierent mobile agent sizes.

0 0.5 1 1.5 2 2.5 3 3.5 1 50 100 150 200 250 Response Time (s)

Number of Managed Network Elements SNMP MA-1kB MA-3kB MA-5kB MA-7kB MA-9kB

Figure 5: Response time for dierent mobile agent sizes.

3.3 Eect of the Task to Be Performed

An analysis related to mobile agent and SNMP's behavior for dierent tasks is performed. Three tasks are used: task t

1 is the same task used in previous sections task t

2 is related to

the retrieval of the SNMP variable named sysContact, which is the name of contact person and how to contact them and task t

3 is related to the SNMP variable called sysDescr, which is a

textual description of the system. Request and response packet sizes for the tasks, including all overhead (PDU GetRequest/GetResponse, UDP, and IP headers) are in Table 2. The link lvalue is the same of the previous experiment, and the mobile agent size is 5 kbytes.

Figure 6 shows that the number of bytes varies with the task to be performed because the greater the number of bytes of every request/response, the larger the total number of bytes transmitted

Table 2: Request and response packet sizes for the tasks. Task Request (bytes) Response (bytes)

t 1 71 76 t 2 69 91 t 3 66 173 0 10 20 30 40 50 60 1 50 100 150 200 250 Number of kbytes

Number of Managed Network Elements SNMP-Task t1 SNMP-Task t2 SNMP-Task t3 MA-Task t1 MA-Task t2 MA-Task t3

Figure 6: Number of bytes for dierent tasks.

0 1 2 3 4 5 6 1 50 100 150 200 250 Response Time (s)

Number of Managed Network Elements SNMP-Task t1 SNMP-Task t2 SNMP-Task t3 MA-Task t1 MA-Task t2 MA-Task t3

Figure 7: Response time for dierent tasks.

and received by the management station. All mobile agent curves start at 10 kbytes since agents traverse the bottleneck link two times and their initial size is 5 kbytes. In Figure 7, response time varies in a dierent way when using the mobile agent or the SNMP. For the SNMP, the task does not have a great inuence due to the fact that the number of bytes exchanged between

the management station and the MNEs is small but for the mobile agent, as the number of exchanged bytes increases (e.g., from task t

1 to task t

3), the response time also increases since

the mobile agent will move with higher diculty.

3.4 Eect of Returning to the Management Station

It has been shown in the previous sections that the mobile agent size increases with the number of visited nodes and, as a consequence, migration becomes dicult. In this section, we evaluate the performance gain of the strategy of returning to the management station when the mobile agent size becomes large. The simulations consider that the mobile agent visits a xed number of nodes per trip, i.e., the number of visited MNEs before returning to the management station to unload. The considered tasks are the three previous ones but their purpose is to retrieve the variables from all 250 MNEs. The link l value and the mobile agent size are the same of the previous experiment. The number of visited nodes per trip varies from 1 to 250.

0 2 4 6 8 10 12 14 1 50 100 150 200 250 Response Time (s)

Number of Elements Managed per Trip MA-Task t1 MA-Task t2 MA-Task t3

Figure 8: Response time when returning to the management station.

Figure 8 shows that the response time decreases sharply when the number of elements managed per trip is small since the agent visits few nodes and returns to the management station. As the number of nodes visited per trip goes on increasing, response time decreases up to a point when it starts to increase due to agent migration diculty related to the agent size. The optimum point varies with the task for tasks t

1 and t

2, visiting 50 nodes and then returning to the

management station provides the best result, while for taskt

3, 25 is the number of nodes to be

visited per trip. For the number of bytes, it decreases when the number of elements managed per trip increases, as in Figure 9, because the more the mobile agent returns to the management station, the larger the number of bytes transmitted and received by this station.

0 500 1000 1500 2000 2500 3000 1 50 100 150 200 250 Number of kbytes

Number of Elements Managed per Trip MA-Task t1 MA-Task t2 MA-Task t3

Figure 9: Number of bytes when returning to the management station.

4 Conclusion

This paper has analyzed the scalability of mobile agents in network management tasks. The performance of mobile agents has been compared with the SNMP one in several simulations. Response time results show that the mobile agent is less sensitive to the latency and the bandwidth of the bottleneck link that connects the management station to the remote LAN, due to the fact that SNMP packets traverse oftentimes the bottleneck link, but it is more inuenced by the task to be performed, since, for the SNMP, the number of bytes exchanged between the management station and the managed network elements is small. The results also show that the mobile agent performs better than the SNMP when the number of managed network elements ranges between two limits: an inferior bound, associated with the number of messages that traverse the bottleneck link, and a superior bound, related to the incremental size of mobile agent, which turns migration dicult. Moreover, the mobile agent performance increases when the agent returns to the management station, after visiting a xed number of nodes. For the number of bytes, the mobile agent performs better than the SNMP when the number of managed network elements exceeds a value related to the overhead of several SNMP retrievals.

We conclude that the mobile agent paradigm signicantly improves the network management performance when most of the agent movement is inside high bandwidth LANs in other words, when only a small percentage of the agent movement concerns the management station and wide area links that are bottlenecks.

References

1] W. Stallings, SNMP and SNMPv2: The infrastructure for network management, IEEE Communications Magazine, vol. 36, no. 3, pp. 3743, Mar. 1998.

2] Y. Yemini, The OSI network management model, IEEE Communications Magazine, vol. 31, no. 5, pp. 2029, May 1993.

3] V. A. Pham and A. Karmouch, Mobile software agents: An overview, IEEE Communi-cations Magazine, vol. 36, no. 7, pp. 2637, July 1998.

4] G. Goldszmidt and Y. Yemini, Delegated agents for network management, IEEE Com-munications Magazine, vol. 36, no. 3, pp. 6670, Mar. 1998.

5] A. Bieszczad, B. Pagurek and T. White, Mobile agents for network management, IEEE Communications Surveys, vol. 1, no. 1, Sept. 1998.

6] M. Baldi and G. P. Picco, Evaluating the tradeos of mobile code design paradigms in network management applications, in Proceedings of the 20th International Conference on Software Engineering (ICSE'98), Kyoto, Japan, pp. 146155, Apr. 1998.

7] A. Sahai and C. Morin, Towards distributed and dynamic network management, in Pro-ceedings of the IEEE/IFIP Network Operations and Management Symposium (NOMS'98), New Orleans, USA, Feb. 1998.

8] M. G. Rubinstein and O. C. M. B. Duarte, Evaluating the performance of mobile agents in network management, in Proceedings of the IEEE Global Telecommunications Conference GLOBECOM'99, Rio de Janeiro, Brazil, Dec. 1999.

9] M. Kahani and H. W. P. Beadle, Decentralised approaches for network management, Computer Communications Review, vol. 27, no. 3, pp. 3647, July 1997.

10] S. Waldbusser, Remote network monitoring management information base. RFC 1757, Feb. 1995.

11] H. S. Nwana, Software agents: an overview, Knowledge Engineering Review, vol. 11, no. 3, pp. 140, Sept. 1996.

12] K. McCloghrie and M. Rose, Management information base for network management of TCP/IP-based internets: MIB-II. RFC 1213, Mar. 1991.

13] K. Fall and K. Varadhan, NS Notes and Documentation, tech. rep., The VINT Project, Jan. 1999.